Clearly strings are going to be broken if this is merged.


dfc660c#diff-a16de2a58a10f14a7fc8512f47bbd467869339a6135046bb0274d2fce1151f93R1901

This PR brings back to track.

Current: 

![image](https://github.com/user-attachments/assets/8a89b863-2e48-49c4-992a-1c06fd358fde)
After:

![image](https://github.com/user-attachments/assets/d244aedb-6ec3-4183-81d8-b61ef9e69445)


This should be covered by unit test. I am preparing it in
https://github.com/Cosifne/roslyn/tree/dev/shech/UnifiedSettingsAdvPage,
will cover these strings will that is ready

Roslyn side of dotnet/razor#10744

…ement

…76375)

This is a small allocation, only showing up as about 0.1% of allocationsi n the typing scenario in the csharp editing speedometer test. However, the change is fairly simple, and I prefer the change as it passes around ImmutableArrays instead of IReadOnlyLists.

…ite warnings (#76347)

* Do not bail generating base type initializer in the presence of use site warnings

We were returning early from generating a call to the base record copy constructor initializer if use site diagnostics reported any kind of diagnostic, including warnings. This isn't good in general, but is particularly bad for warnings like CS1701, which are typically suppressed by the SDK and results in what seems like a completely clean csc invocation skipping calling the base constructor. Fixes #72357.

* Set FailsPEVerify

* Only run test on coreclr

…oAssemblyInUsingNamespaces (#76370)

This method shows up as 1.1% of allocations in the typing scenario in the csharp editing speedometer test. Of that, 0.6% is in the string.Concat call merging the Namespace, '.', and type name.

Instead of creating and passing this merged string to GetForwardedToAssembly, we can instead construct a MetadataTypeName to pass to that method. That struct can either be constructed using a fullname or a namespace/typename combination.  In the caller where we are trying to get rid of the string concat, we can pass in the namespace/typename and avoid the concat operation completely.

The other caller to GetForwardedToAssembly also now passes in a MetadataTypeName, but it's constructed from the fullname as that method used to do itself before taking in the MetadataTypeName.

…ry in SolutionCompilationState.ComputeFrozenSnapshot (#76361)

* *** draft PR while I wait on speedometer to verify if this improves allocations ***

Reduce allocations from doing repeated adds into an ImmutableDictionary in SolutionCompilationState.ComputeFrozenSnapshot

Instead, use an ImmutableSegmentedDictionaryBuilder to do the adds and then do a single update to the SolutionCompilationState._projectIdToTrackerMap after all new compilation trackers are created.

Note that there is a potential negative aspect to this change as it increases the time window during which a compilation state can be queried for in the SolutionCompilationState and not found. However, I don't believe this will cause functional issues, and wasn't hitting that concern during local testing.